Hello All,
I have just unexpectedly found an old Beagleboard B7 in my closet and tried to compile the current U-Boot master branch for it. Appears that it is broken since almost 2 years ago:
https://lists.denx.de/pipermail/u-boot/2015-July/220332.html https://lists.denx.de/pipermail/u-boot/2016-May/253777.html
And after trying to narrow it down, looks like the failure happens in the prcm_init() function from 'arch/arm/mach-omap2/omap3/clock.c' when it is compiled as Thumb2. At least the SPL can successfully boot after the following change (GCC6 is needed for this pragma):
diff --git a/arch/arm/mach-omap2/omap3/clock.c b/arch/arm/mach-omap2/omap3/clock.c index 006969e..8064fa6 100644 --- a/arch/arm/mach-omap2/omap3/clock.c +++ b/arch/arm/mach-omap2/omap3/clock.c @@ -592,6 +592,8 @@ static void iva_init_36xx(u32 sil_index, u32 clk_index) wait_on_value(ST_IVA2_CLK, 1, &prcm_base->idlest_pll_iva2, LDELAY); }
+#pragma GCC target ("arm") + /****************************************************************************** * prcm_init() - inits clocks for PRCM as defined in clocks.h * called from SRAM, or Flash (using temp SRAM stack). @@ -700,6 +702,8 @@ void prcm_init(void) sdelay(5000); }
+#pragma GCC target ("thumb") + /* * Enable usb ehci uhh, tll clocks */
Now it's very interesting to figure out what exactly is wrong there. OMAP3530 and DM3730 have different code paths in this function, so it's not very surprising that only OMAP3530 is broken.
Yes, I know that OMAP3530 has a very old and buggy Cortex-A8 core. But I tried to apply all the Thumb2 related errata workarounds early in 'arch/arm/cpu/armv7/start.S'. And also enforced single-issue execution (bit 10 in the AUXCR) as a workaround for the most nasty erratum (657417: A 32-bit branch instruction that spans two 4K regions can result in an incorrect operation) just because I don't fully trust the linker.